The Year in AI Papers
The year/Independent research

Paper 2606.25041

Wan-Streamer v0.1: End-to-end Real-time Interactive Foundation Models

Lianghua Huang, Zhi-Fan Wu, Wei Wang, Yupeng Shi, Mengyang Feng, Junjie He, Chen-Wei Xie, Yu Liu, Jingren Zhou, Ang Wang, Bang Zhang, Baole Ai, et al.

Published
Jun 2026
Research lab
Independent
Citations
4
GitHub
Not linked

01 In brief

Summary

Wan-Streamer is a native-streaming, end-to-end interactive foundation model from Alibaba Group designed for real-time, low-latency, full-duplex audio-visual interaction.

It models language, audio, and video as both input and output within a single Transformer, using block-causal attention for incremental streaming.

Unlike cascaded systems, it does not rely on external VAD, ASR, language, TTS, or video-generation modules; perception, reasoning, generation, and turn management are learned jointly.

The architecture includes strictly causal audio and video VAEs, causal encoders/decoders, and a block-causal Transformer, enabling streaming units as short as 160 ms at 25 fps.

Inference uses a thinker-performer pipeline with KV-cache exchange to overlap perception, generation, and decoding, achieving approximately 200 ms model-side response latency and approximately 550 ms total interaction latency (including 350 ms network).

Training occurs in three stages: independent-task pretraining, end-to-end interaction training, and distillation for low-latency streaming.

Experiments show Wan-Streamer outperforms or matches existing systems in latency and naturalness, supporting sub-second duplex communication with synchronized visual output.

Current v0.1 results are validated at 192p resolution, with scaling to higher resolutions left for future work.

02 From the paper

Abstract

We present Wan-Streamer, a native-streaming, end-to-end interactive foundation model designed from the ground up for real-time, low-latency, full-duplex audio-visual interaction. Wan-Streamer seamlessly models language, audio, and video as both input and output within a single Transformer, where the sequence is represented as interleaved visual, audio, and text input tokens together with visual, audio, and text output tokens, coordinated by block-causal attention for incremental streaming. Unlike cascaded interactive systems that rely on separate VAD, ASR, language, TTS, audio-driven animation, or video-generation modules, Wan-Streamer does not rely on external language, speech, avatar, or video-generation modules: perception, reasoning, generation, response timing, turn management, and cross-modal synchronization are learned jointly within one unified model, reducing pipeline latency and error accumulation. To support natural audio-visual responsiveness, we redesign the entire stack around streamability, including causal encoders, causal decoders, block-causal attention, and low-latency multimodal token scheduling, enabling streaming units as short as 160 ms at 25 fps. Wan-Streamer achieves approximately 200 ms model-side response latency and approximately 550 ms total interaction latency when combined with 350 ms bidirectional network latency, supporting sub-second duplex audio-visual communication. These results position Wan-Streamer as a unified, end-to-end, multimodal interactive foundation model for low-latency streaming interaction.